Data Center Capacity Planning Calculator
Project when power, space or cooling capacity runs out based on current utilisation and compound growth rate.
Inputs
Whatever you are tracking: kW, racks, U, TB. Keep units consistent.
Compound growth per year, measured from your own history.
Utilisation at which you must start procurement, not finish it.
Results
- Time until the warning threshold
- 1.6 years
- Time until capacity is exhausted
- 2.8 years
- Current utilisation
- 60.0%
- Headroom remaining
- 200.0
- Projected consumption in 12 months
- 360.0
Start procurement at this point, allowing for lead times.
The projection
Capacity consumption compounds. Rearranging the compound growth formula to solve for time gives:
years = ln(target ÷ current) ÷ ln(1 + growth rate)
At 300 kW of a 500 kW facility growing 20% a year, you cross 80% utilisation in 2.7 years and run out entirely in 2.8. That gap — barely a month between "getting full" and "full" — is the point of the exercise. Compound growth compresses your warning period as you approach the limit.
Why 80% is the number that matters
The date capacity runs out is not the date you need to act. Working backwards from the threshold:
- Procurement and approval: 1-3 months for budget sign off in most organisations.
- Equipment lead time: 3-6 months, and considerably longer for electrical infrastructure like switchgear, transformers or UPS. This has been the binding constraint across the industry recently.
- Installation and commissioning: 1-3 months, plus load testing.
That is comfortably a year, often more. If the calculator says you hit 80% in nine months, you are already late. Set the threshold to whatever gives you your real lead time, not to a round number.
Track every dimension separately
A facility does not run out of "capacity" — it runs out of one specific thing, and usually not the one being watched:
| Dimension | Runs out when | Tool |
|---|---|---|
| Power | Usually first, in modern deployments | PDU load |
| Cooling | Often at a lower density than power allows | Cooling load |
| Space | Rarely the first constraint any more | Rack units |
| Weight | With dense storage on older floors | Floor loading |
| Network ports | Quietly, between refresh cycles | — |
Run this projection once per dimension. The earliest date is your real planning horizon, and it is frequently power.
Where the model is wrong
Smooth compound growth is a convenient fiction. Real consumption is lumpy:
- Demand arrives in steps. A single project can consume a year of projected growth in one deployment weekend.
- Refresh cycles cut both ways. New hardware is denser per rack but often more efficient per unit of work, so power can fall while capability rises.
- Stranded capacity is invisible here. A facility at 60% overall can still be unable to accept a new rack, because the headroom is distributed as 2 kW in each of thirty cabinets rather than 60 kW in one place.
- Growth rates are not stable. Derive yours from at least two years of your own measurements. An industry average is not your number.
Use it as a trigger, not a forecast
The value of this calculation is not the precision of the date. It is that it converts a vague sense of "we are filling up" into a specific month when procurement must begin — and that it makes the compression near the limit visible before you are living in it.
Re-run it quarterly with updated numbers. A projection made once and filed is worth very little.